Kognitions- og kunstig intelligens-forskning mellem logik og psykologi: et semiotisk synspunkt

Artiklen præsenterer et »meta-semiotisk« perspektiv på kognitionsforskningen (CS), ifølge hvilken man hverken bør søge grundlaget for CS i den psykologiske udforskning af menneskelig intelligens eller i konstruktionen af en »kunstig intelligens« (Al), men derimod i udforskningen af de mulige udformninger af kognitive processer og formationer. Teoretisk Al vil understøtte en sådan forskning ved at analysere egenskaberne for datamatiske modeller, der kan simulere »det kognitive«.CS bør interessere sig specielt for biologisk implementeret intelligens, men det bør ske på baggrund af et teoretisk arbejde med forskellige niveauer af »tvangsbindinger« (logiske, fysiske, biologiske etc. constraints), ud fra hvilke vi kan erkende de forskellige naturlige og kunstige implementeringer af intelligens som realiseringer af abstrakte muligheder, der knytter sig til tegn, redskaber og modeller. Metasemiotikken studerer de almene egenskaber ved tegn (og ved betydning og kommunikation), redskaber, herunder maskiner) og modeller (fysiske og mentale modeller). CS står over for tre fundamentale problemer: I) at overvinde modstillingen af psykologiske og logiske opfattelser af kognition, 2) at reformulere den fænomenale verdens status inden for teorier om kognition og 3) at reformulere konflikten mellem det klassiske Al-paradigme og det konnektionistiske paradigme (»neuralenetværk«) som et index på et reelt teoretisk problem, - nemlig spørgsmålet om forholdet mellem symbolske og sub-symbolske processer og om integrationen af diskrete og kontinuerte modeller af disse kognitive processer.The article presents a »meta-semiotic« perspective on cognitive science (CS) according to which the foundation of CS should be sought, neither in the psychological investigation of human intelligence nor in the »frankensteinian« construction of an artificial intelligence, but in the investigation of possible cognitive formations. Theoretical research in artificial intelligence would support such an investigation by analyzing the properties of computational models that can simulate these cognitive formations. CS should take a special interest in the biological implementation of cognitive formations (in order to enhance the scientific understanding of the mental faculties), but I propose that it should be done on the background of theoretical research into differentlevels of constraints (logical, physical, biological etc.), that can help us recognize the different (natural or artificial) implementations of »intelligence« as realizations of abstract possibilities relating to the reality of signs, tools, and models. The general properties ofsigns (as well as the general properties of signifieation and communication), the general properties of tools (as well as abstract machines), and the general properties of models (physical and mental), - is the subject of meta-semiotics. CS is at present confronted by three fundamental problems: 1) to overcome the dichtomy between psychological and logical conceptions of cognition, 2) to reformulate the status of the phenomenal world in theories of cognition (thc status of »naive« and commonsensetheorics), and 3) to restate the conflict between the paradigm of traditional AI and the paradigm of ncural networks as an index of a reel theoretical problem, by formulating possible integrations of discrete and continuous models of symbolic and subsymbolicprocesses